Key Points from Jensen Huang’s GTC Speech: Revenue Forecast Raised to $1 Trillion, Launch of Minimalist Shrimp-Farming NemoClaw Agent Platform

On March 17, early Tuesday morning in the East 8th District, NVIDIA CEO Jensen Huang delivered a “carpet bombing” of the AI industry's software and hardware concepts in a two-and-a-half-hour-long keynote. For the capital markets, it was also a day of substantial gains—almost all the hyped concepts anticipated in advance were realized, and unexpectedly, Jensen Huang provided explosive new financial guidance on computing power chip revenue.

1. From a chip company to a $1 trillion “Token Factory”—Economic Restructuring in the Era of Inference.

Jensen Huang announced that computing demand has entered a stage of “million-fold growth,” doubling the prediction for computing power needs from $500 billion to $1 trillion for the period up to 2027. He defined the concept of a “Token Factory,” indicating that future data centers will not be storage centers anymore, but rather factories producing intelligent tokens.

2. Vera Rubin Architecture and Groq Integration: Decoupled Inference Unlocks 350x Acceleration.

The keynote unveiled the next-generation architecture codenamed Vera Rubin. The system features a fully liquid-cooled design and introduces Vera CPU, optimized for agent-based tasks. The most significant breakthrough is the integration of Groq team’s deterministic stream processor technology. Through Dynamo software, they achieved “decoupled inference”: using Rubin’s massive memory to handle complex contexts, while employing Groq chips to generate tokens at blazing speed. This combination enables token generation speeds in gigawatt-scale factories to achieve a staggering 350-fold leap forward.

3. OpenClaw: The “Linux Moment” for AI and Intelligent Agent Operating System.

Jensen Huang described the open-source project OpenClaw as the Linux of the AI era. It is an OS-level framework that allows AI agents to autonomously call tools, execute code, and manage file systems. To address enterprise security pain points, NVIDIA introduced the minimalistic, ready-to-use NemoClaw agent platform, integrating the OpenShell privacy guardrail—signaling a comprehensive industry shift from SaaS (Software as a Service) to AaaS (Agent as a Service).

4. The “ChatGPT Moment” for Physical AI: From Autonomous Driving to Disney’s Olaf.

The keynote showcased the explosive growth of physical AI, announcing that autonomous driving has entered mass commercialization. The NVIDIA RoboTaxi Ready platform added partners such as BYD, Hyundai, Geely, and others, covering 18 million new cars annually, and has reached a deployment agreement with Uber. The highlight was the appearance of Disney’s robot Olaf, demonstrating physical adaptability evolved within virtual worlds through Omniverse and the Newton solver.

5. Feynman Architecture and Space Plans: Vertical Integration Toward the Stars and the Sea.

Looking ahead, Jensen Huang unveiled Feynman, the next-generation GPU architecture following Rubin, which will feature LP 40 processors and Rosa CPUs, supporting CPO (Co-Packaged Optics) technology. In addition, NVIDIA announced its expansion into space computing, developing the Vera Rubin Space-1 computer. Through the Omniverse DSX platform, the company is building a global (and even space-based) digital twin design system, achieving vertical integration from the hardware layer to cosmic-scale infrastructure.